Google: Alleviating Climate Change Impacts With Physics & AI

According to Google, precipitation remains one of the trickiest tasks for global-scale weather and climate models, due to unknown times, amounts and locations.

Google is introducing its open-sourced hybrid atmospheric model, NeuralGCM, combining machine learning (ML) with physics to run fast and accurate global atmospheric simulations.

NeuralGCM more accurately reproduces average precipitation, extreme rainfall and the daily weather cycle, delivering particularly strong improvements for the most intense 0.1% of rainfall events. 

Developed as part of the wider Earth AI initiative, the hybrid physics and AI model complements AI-only systems such as Google’s recently updated WeatherNext 2 by extending capabilities for longer-range weather and climate analysis.

Using AI for Physics-related weather

According to the UK Met Office, climate change can lead to an increase in melting sea ice/glaciers, rising sea levels, earths hydrological cycle and both an increase in heatwaves and rainfall.

In order for Google to simulate precipitation it needs to monitor clouds; however, these come in different sizes, types and characteristics, making it hard for large-scale models to resolve issues.

To account for the effect of small-scale atmospheric processes like cloud formation on the climate, models use approximations called parameterisations, which are based on other variables. 

Instead of depending on these parameterisations, Google’s NeuralGCM uses a neural network, learning the effects of such small-scale events directly from existing weather data.

In this version of NeuralGCM, precipitation is represented more accurately by training the machine learning component directly on satellite-based precipitation observations rather than relying on reanalyses. 

Earlier iterations of Google’s model, like most machine learning weather systems, were trained on reconstructed atmospheric conditions that blend physics-based models with observations, a process that often struggles to capture the complexity of cloud physics and, in turn, precipitation extremes and daily cycles. 

To overcome this, the precipitation module was trained on NASA satellite observations collected between 2001 and 2018, enabled by NeuralGCM’s differential dynamical core. 

Unlike previous hybrid physics and AI models, which were limited to reanalysis or simulation outputs, this approach allows NeuralGCM to learn a more accurate, machine-learned parameterisation of precipitation directly from high-quality observational data.

“Better weather models equal better climate resilience,” writes Robert Little, Sustainability Strategy Lead & Subject Matter Expert at Google, on LinkedIn.

“I’m excited to see where this takes the field of climate science next.”

How can Google forecast precipitation?

NeuralGCM’s performance was assessed using WeatherBench 2 across two-week forecasts and compared with a leading physics-based model from the European Centre for Medium-range Weather Forecasts (ECMWF). 

Using forecasts initialised at noon and midnight throughout 2020, with data excluded from training, NeuralGCM outperformed the ECMWF model at low resolution across most precipitation metrics, including both 24-hour and 6-hour accumulated rainfall over all 15 forecast days, with particularly strong results over land where impacts on people and ecosystems are greatest. 

Although its current 280 kilometre resolution is too coarse for operational forecasting, the results indicate clear potential for applying this approach at finer scales. 

Over longer timescales spanning years to decades, NeuralGCM also demonstrated strong skills in: 

• Reproducing average and extreme precipitation patterns
• Achieving an average mean error of less than half a millimetre per day
• Reducing error by 40%  compared with leading global atmospheric models used in the latest Intergovernmental Panel on Climate Change report, with even larger improvements over land. 

The model showed marked gains in capturing extreme rainfall events, particularly the most intense 0.1% of precipitation, addressing long-standing issues in physics-based models that tend to overestimate light rainfall while underestimating heavy events. 

NeuralGCM also more accurately reproduced the daily timing and intensity of precipitation, including strong diurnal cycles such as afternoon rainfall in the Amazon during summer, outperforming traditional climate models that often produce rain too early in the day. 

Accurately capturing when and where precipitation falls is critical for applications ranging from flood and drought management to climate science, ecosystem resilience and public safety.

The future of AI

“We believe this is a step forward for large-scale precipitation forecasts and simulations, and we already have early support in the real world,” says Google.

According to the Intergovernmental Panel on Climate Change (IPCC)’s Sixth Assessment Report released in 2021, the human-caused rise in GHGS has increased the frequency and intensity of extreme weather events like monsoons, droughts and heatwaves.

A partnership between the University of Chicago and the Indian Ministry of Agriculture and Farmers Welfare used Google’s NeuralGCM to predict the onset of the monsoon season.

In summer 2025, the pair selected NeuralGCM and one other model to build and deploy a forecasting tool.

“Since introducing NeuralGCM we have made everything available as open-source code on which we hope people can build,” says Google.

“This precipitation model is also being openly released to the extended community.

“Ultimately our hope is that these efforts will bring us one step closer to accurate long-term projections of future precipitation, especially under climate change.”